Torsion sections of elliptic surfaces - Tsinghua...

Ark Mat 33 (1995), 117-134 ~) 1995 by Institut Mittag-Lei~er All rights reserved

Torsion sections of elliptic surfaces

Rick Miranda( 1 ) and Peter Stiller

Abstract Given a torsion section of a semistable elliptic surface we prove equidistribution results for the components of singular fibers which are hit by the section and for the root of unity (identifying the zero component with C) which is hit by the section in case the section hits the zero component

1 I n t r o d u c t i o n

In this article we discuss torsion sections of seml.qtable elliptic surfaces defined

over the complex field C Recall tha t a semistable elliptic surface is a fibration ~r: X---~C, where X is a smooth compact surface, C is a smooth curve, the general fiber of ~r is a smooth curve of genus one, and all the singular fibers of Ir are

seml.qtable, tha t is, all are of type Im in Kodaira ' s notat ion (see [K]) In addition, we assume tha t the fibration ~r enjoys a section So; this section defines a zero for a

group law in each fiber, making the general fiber an elliptic curve over C

The MordeU-Weil group of X , denoted by M W ( X ) , is the set of all sections of

r , which forms a group under fiber-wise addition; the section So is the identity of M W ( X ) Note tha t any section SEMW(X) meets one and only one component of each fiber

Given any section SEMW(X), one can ask the following two questions First, which components of the singular fibers of X does S meet, and second, exactly where in these components does S meet them?

When S is a torsion section, the first question was addressed in [M] To describe those results we require some notat ion Recall tha t a singular semi.~table fiber of type Im is a cycle of m P l ' s Suppose tha t our elliptic fibration ~r: X---*C has s such singular fibers F1, , Fs, with Fj of type Im~ Choose an "orientation" of each

(1) Research supported in part by the NSF under grant DMS 9104058 and the NSA under grant MDA 904-92 H 3022

118 R i c k M i r a n d a a n d P e t e r S t i l l e r

fiber Fj and write the mj components of Fj as

C 0 ( J ) ~(/) r,(J) ' V l ~ , ~ J m j - - 1

where the zero section So meets only C0 (j) and where for each k, Ck (j) meets only Ck(~X

rood mj If rnj=l, then Fj=C (j) is a nodal rational curve with self-intersection 0

If mj>_2, then each C (j) is a smooth rational curve with self-intersection - 2 Given a section S of X, and orientations of all of the singular fibers Fj, inducing

a labeling of the components as above, we define "component numbers" kj (S) to be the index of the component in the jth fiber Fj which a given section S meets That is,

S meets Ckj(/)(s) in the fiber Fj

This assigns to each section S, an s-tuple (kl(S), , ks(S)) The component number kj can be taken to be defined rood mj once the orientation of Fj is chosen; if the orientation is unknown, then kj is defined rood rnj only up to sign Note that kj(So)=O for every j; indeed, after choosing orientations, the assignment of component numbers can be considered a group homomorphism from MW(X) to Sj z/mjz

Now suppose that S is a torsion section of prime order p In this case, since p S=So, we must have pkj(S)-O mod mj for every j If p does not divide mj, then this forces kj(S)=O However if mj=pnj, then kj(S) can a priori be any one of the numbers inj, for i=0, , p - 1 This multiple i measures, in some sense, how far around the cycle S is from the zero-section So in the jth fiber

Of course, changing the orientation in the fiber Fj will have the effect of changing this multiple i to p - i (if i#0; i=O rem~in.q unchanged) We are therefore led to a definition of the following quantities Let Mi (S) denote the number of singular fibers where kj(S)=inj or kj(S)=(p-i)nj (weighted by the number mj of components in the fiber Fj), and then divided by the total number Y~j mj of components:

M , ( S ) =

mj j with kj(S)=inj or (p-i)nj

~-~ mj J

We may view Mi as a probability, since it is the fraction of the fibers where S meets "distance i" away from the zero section

The main result of [M] is the following:

Torsion sections of elliptic surfaces 119

T h e o r e m 1 1 1/S is a torsion section of odd prime order p, then M~(S)= 2p/(p2-1) if leO, and Mo(S)=I/(p+ I)

Notice that, firstly, these fractious are independent of S, and secondly, that they are independent of i (for i non zero) Thus we obtain an "equidistribution" property for these component numbers The proof of the above theorem given in [M] used only basic facts about elliptic surfaces and some intersection theory

In this paper we will address the second of the two questions raised above in a similar spirit Again let S be a torsion section of order p, and consider those fibers where kj(S)=0, i e , where S and So meet the same component By the above result, this happens exactly 1 / (p+l ) of the time, where each fiber is weighted by the number mj of components it has Each identity component C0 (j) may be naturally identified as a group with C*, by sending the two points of intersection with the neighboring components to 0 and co, and the point where So meets C~ ) to 1 (This identification can be made in exactly two ways, corresponding to which node is sent to 0 and which to oo ) Given such a coordinate choice on CO(j), the section S of order p will hit C~ ) at a point whose coordinate is a pth root of unity

Let us denote by lj(S) that integer in [1 ,p- l ] such that the point SNCO(j) has coordinate exp(27rilj(S)/p) (Since torsion sections can never meet, lj(S) cannot equal 0 ) Note that lj(S) is defined only when k j (S)=0 and may be thought of as being defined modulo p; in addition, if we switch the roles of 0 and c~, we see that lj(S) is replaced by p- l j (S) Thus, combinatorially, we are in a situation identical to the one for the component numbers kj We will call these numbers lj (S) root of unity numbers for the section S

Now define numbers R/(S) to be the total number of fibers having kj(S)=O and having lj(S)=4-i mod p (weighted by the number of components rnj), and then divided by the total (weighted) number of fibers with kj (S) =0:

P~/(S) = j with kj(S)=0 and lj(S)=:l=i

mj j with kj(S)=O

The main theorem of this article is an equidistribution property for these fractious R/(S): namely, i fp is odd, then P~(S)=2/ (p-1) , independent of S and i

These results rely on computations on elliptic modular surfaces We also give as a by product the equidistribution for the component numbers kj(S), both by an explicit computation and via a relatiouship between the component numbers and the root of unity numbers Finally we develop an Abel theorem for a singular semistable elliptic curve and use it to compute a Weil pairing on the singular semistable fibers

120 Rick Miranda and Peter Stiller

of an elliptic surface Using this Weil pairing, we may also discover the duality between the component numbers and root of unity numbers

2 Equidistribution for the roots of unity via the universal p rope r ty

In this section we will give a proof of the following theorem

T h e o r e m 2 1 Fix a prime number p, and let S be a section of a smooth semistable elliptic surface which is torsion of order p Then

R I ( S ) = I i f p=2 , and P~(S)=2 / (p -1 ) i fp is odd

Proof First note that if p---2, then the only possible value for lj is one, so that certainly RI (S )=I Similarly, if p=3, the only values for lj are 1 or 2, which are inverse rood 3; therefore RI(S )=I again Thus we may assume that p is an odd prime at least five

Second, note that it is immediate to check that if S is a section of 7r: X-*C , and if F: D-*C is an onto map of smooth curves, then S induces a section S ~ on the pull-back surface 7r~: X~-*D, which is also torsion of order p Moreover, it is easy to see that P~(S')=R~(S) for every i

If p_>5, then we may consider the elliptic modular surface YI(P) over the modular curve X1 (p), which is defined using the congruence subgroup F1 (p) of SL(2, Z) given by

Note that X1 (p) is a fine moduli space for elliptic curves with a torsion point of order p, and that YI(p)--*XI(p) is the universal family (see [Shm], [Shd] or [CP]) This elliptic modular surface also has a universal section T of order p, and every elliptic surface with a section S of order p may be obtained via pull-back from this modular surface in such a way that S is the pullback of T By the above remark concerning the constancy of the fraction R~ under pull-back, it suffices to prove that

1 1 /~ (T)=2 / (p -1 ) , for each i=l , , ~ (p- ); in other words, we need only verify the statement of the theorem for the universal section T of the modular surface Y1 (P) over the modular curve X1 (p)

Now the elliptic modular surface has exactly p - 1 singular fibers, occuring over the cusps of the modular curve X1 (p), of which half are of type /1 and half of type Ip depending on the order of the cusp: the total '~veight" of the singular fibers, which is equal to the Euler number of the modular surface, is ~'~j m j = �89 (p2_ 1) For the/1 fibers, the component numbers kj (T) must be 0, contributing


[ �89189 to the weighted fraction Mo(T) However, by the results of [M] mentioned in the introduction, the total weighted fraction Mo(T) is 1/(p+l) , and therefore the component number must be non-zero for all the fibers of type Ip

At this point, label the /1 fibers as F1, ,F(v_l)/2 Choose an "orientation" of each/1 fiber, which in essence means give an isomorphism of its smooth part with C*, such that the zero section To meets Fj in the point corresponding to the number 1

Denote by TI=T, T2=2T, ,Tp_l=(p-1)T the non-zero multiples of the uni versal section T Note that T~ and Tv_i meet each 11 fiber Fj in inverse roots of unity; in other words, using the notation of the Introduction, lj (Ti)=-lj (Tp-i) mod p for every i and j Construct a square matrix Z of size �89 whose ij th entry is the pair of integers q-lj(Ti)={lj(Ti), lj(Tp_i)}

Now two torsion sections of an elliptic surface never meet, so the sections {Ti} are all disjoint In particular, if we fix an index j , in the jth column of this matrix Z, we must have p - 1 distinct integers; since the integers come from [1,p-1], each integer in this range occurs exactly once in each column of Z

Next fix an integer ie [1, x ~(p-1)] By the universality of the modular surface, there is an automorphism of the surface fixing the zero section and carrying Ti to T=T1 Therefore, the entries along the ith row must be the same as the entries along the first row, but permuted in some way (indeed, permuted as the automorphism permutes the Ix fibers Fj)

As noticed above, each integer in [1,p-1] appears exactly once in the first column of Z Hence each integer appears in some row of Z By the above remark, each integer must then appear in the first row, and indeed in every row This is exactly the statement that R/ (T)=2/ (p-1) in this case []

3 Equidistribution for the roots of unity via explicit computat ion

In this section we will re prove Theorem 2 1 via an explicit computation Fix an odd prime p>5, and let F=FI(p) be the relevant modular group defined in the proof of the theorem Let H denote the upper half-plane Form the semi-direct product F = F o Z 2 and recall that F acts on H x C by

(aT+b

Denote the quotient H x C / r by yO(p), and the quotient H/r by X~ The natural map r: Y~176 induced by sending (T, w) to r is a smooth elliptic fibration

122 Rick M i r a n d a a nd Pe t e r St i l ler

The curve X~ is not compact, but these quotients and the fibration ~ extend to the natural compactifications lr:YI(p)-~XI(p) The compact curve XI(p) is obtained by adding p - 1 points (called cusps) to the open curve X~ over which the elliptic fibration has singular curves These cusps correspond to an equivalence class of rational points on the boundary of the upper half-plane and the vertical point ioo at infinity

For this modular group FI(p), representatives for the cusps can be taken to be

the rational numbers

r 1 1 1 with 1 < r < _--(v-I), and - with 1 < s < : ( p - l )

p ' 2 ~ " s - - 2 - - -

(See [CP] ) The cusp at ioo is equivalent to the cusp l ip Over the cusps of the form r/p, we have singular fibers for ~r of type I1; over

the cusps of the form 1/s, we have singular fibers of type Ip There are exactly p sections of the map ~r, which are induced by letting

= - -

P

for 0 < a < p - 1 (See [Shd] ) The zero-section To for ~r is of course defned by w(r)=0, while the universal section T can be defined by W(T)=I/p

A local coordinate for the modular curve X1 (p) about the point corresponding to the cusp at T=ico is q=exp(2~riT); the fibers of the modular surface itself may be locally represented near this point as C/(Z+Z~)-~C*/q z We want to determine a local coordinate about the point corresponding to the cusp at r=r /p For this, choose b, dEZ such that rd-bp=l , and let

which is then in SL(2, Z) This element 7 sends rip to ioo via the action of (3 1); for our purposes its effect is to induce an isomorphism

C/(Z+ZT) = C / ( Z ( r - p r ) +Z(dT-b)) =~ C*/q z

where this last map sends w to exp (21riw/(r- pr)); here q = exp ( 27ri (dT -- b) / (r - pr ) ) is the local coordinate near ~-=r/p

Since a/p=a(dT-b)+(ad/p)(r-pT) , we see that the section w(v)--a/p maps under the above isomorphism to exp(27ri(a~,(r)+ad/p)) As T approaches r/p, ~(r) approaches ioo and this quantity approaches exp(2~riad/p) This is the root of unity which we have desired to compute

Torsion sect ions of ell iptic surfaces 123

Note that had we used -~/instead of ~/we would have gotten the inverse root of unity, and that d is determined only mod p

Using the notation of the introduction, let us label the /1 fibers of the modular surface as F1, ,F(p-1)/2, with Fr corresponding to the cusp at ~'=r/p We have shown that for the section T~=(~T,

= a t -1,

where the value of r -1 is taken modulo p This gives an alternate proof of Theorem 2 1, since for any fixed a, these values

are equidistributed among the nonzero classes mod p, up to sign

4 Equidistribution for the component numbers via explicit computation

The main theorem of [M] concerning the equidistribution of the component numbers can also be proved by appealing to the universal property of the modular surface 7r: YI(p)--'XI(P) Using the notation of the introduction, one sees that the fractions Mi(S) for a torsion section S of order p remain unchanged under base change Thus it suffices to check that they have the correct values, namely those given by Theorem 1 1, for the universal section T of the modular surface This we do in this section

As noted in the previous section, the modular surface contains the fibers F1, ,Fr, ,F(p-1)/2 of type /1 over the cusps represented by the rational numbers r/p, for l < r < � 8 9 For these fibers, since they have only one component,

1 1 2 the component number k~ (T) is zero, contributing [~ (p - 1)] / [3 (P - 1)] = 1/(p+ 1) to the fraction M0 (T)

The rest of the singular fibers of the modular surface will be denoted by F(p+l)/2, , Fp-1 where Fp-s will be taken as the fiber over the cusp represented by the rational number l / s , for 1 < s < �89 (p - 1 )

The component number theorem then follows from the next computation

Proposition 4 1 With the above notation, if l < s < � 8 9 then kp_s(T)= =l=s (the indeterminacy of the sign being due to the choice of orientation of the singular fiber Fp-s )

Proof It is convenient to again transport the computation to r=ioo as was done in the previous section Fix an s, with l < s < � 8 9 and let

124 PAck Miranda and Peter Stiller

As r approaches 1/s, 7(~')=~-/(1-s'r) approaches ioo

This element 0' induces an isomorphism

c / ( z + z r ) = c / ( z ( 1 - --- C*/q

where this last map sends w to exp(27riw/(1-s~')); here q----exp(27riT/p(1--ST)) is the local coordinate near this cusp

Since 1/p=(1/p)(1--ST)+(s/p)'r, we see that the universal section T, which is defined by w (T) ---- 1/p, maps under the above isomorphism to exp (21ri (1 + s7 (T))/p) ---- CpqS, where Cp=exp(2 i/p)

Now using the toric description of the surface near this cusp, (see [AMRT, Chapter One, Section 4]), the exponent of q (modulo p) in this formula governs which component is hit by the section In particular, the universal section hits component s in the fiber over the cusp represented by 1/s Note that using - 7 instead of -~ would result in - s instead of s []

5 T h e c a n o n i c a l i n v o l u t i o n

The matrix

:) normalizes the congruence subgroup FI(p), and therefore induces an automorphism of the modular curve Xl(iV) Since .4 2 is a constant multiple of the identity, this automorphism (which we will also call A) is an involution, called the canonical involution; in terms of the variable T, .4 takes 7" to - 1 / p r The involution .4 also permutes the cusps, exactly switching the two sets of �89 cusps; indeed, the rational number rip is taken by A to the number -1/r , which is equivalent to the number 1/r under the action of r l(p) Thus the /1 cusp represented by rip is switched via .4 with the Ip cusp represented by 1/r

Hence we observe that under this correspondence, the/1 cusp having u as the root-of-unity number for the universal section T is paired with the Ip cusp having u - I as the component number for T

Note that for the h fibers, all sections have component number kj equal to zero, and all non zero sections have root-of-unity number lj in G(p)=(Z/p)X/~l Moreover, for the Ip fibers, all non zero sections have component number kj also in this value group G(p) For a cusp x of type I1, denote the root-of-unity number of the universal section T in G(p) by lx; for a cusp x of type Ip, denote the component


number of the universal section T in G(p) by kx With this notation the above statement can be re phrased as follows

(5 i) For every cusp x of type /1 , lx -- kA~ in G(p)

This is a manifestation of a certain duality between the two notions This we will explore further in the next sections, via a version of the Weil pairing on singular fibers of elliptic surfaces Before proceeding to this, we want to make some

' elementary remarks concerning the modular surface and this involution As above, denote the modular surface over XI(p) by 7r:YI(p)-*XI(p) Let

r ' :YI(p) ' -- ,XI(p) denote the pull back of ~r via the involution A This operation exactly switches the fibers over the cusps, so that the fiber of ~d over an riP cusp is now of type Ip, and the fiber of 7d over a 1Is cusp is of t ype /1 The universal section T of r pulls back to a section T' of ~' (We note that this surface is the modular surface for the group Fl(p) )

An alternate way of constructing this elliptic fibration ~r' is to take the original modular surface 7r: YI(p)--*XI(p) and divide it, fiber-by-fiber, by the subgroup generated by the universal torsion section T Over the cusps represented by r/p, the original modular surface has I1 fibers, which are rational nodal curves; in the quotient there is again a rational nodal curve, but the surface acquires a rational double point of type Ap-1 at the node, and the minimal resolution of singularities produces a fiber of type Ip Over the cusps represented by l/p, we have Ip fibers in the original surface; in the quotient the p components are all identified to a single /1 component For details concerning this quotient construction, the reader may consult [MP]

The section T' in this view comes not from the original section T, but from a p- section of the modular surface, consisting of a coset of the cyclic subgroup generated by T in the general fiber of ~r (The universal section T and all of its multiples of course descends to the zero-section of 7d ) The formula (5 1) implies that

kx(T') =lx(T) -1

in the group G(p)

6 T h e f u n c t i o n g r o u p o f a s e m i s t a b l e e l l i p t i c c u r v e

In the next section we will develop a version of the Weil pairing on a semistable singular fiber of an elliptic surface (that is, a fiber of type Ira) The essential ingredient in the definition of the Weil pairing on a smooth elliptic curve is Abel's


theorem; see for example [Sil, Chapter III, Section 8] Therefore we require a version of Abel's theorem for a fiber of type Ira, and this in turn requires a notion of appropriate rational functions on the degenerate fiber In this section we describe the set of functions which we will use; Abel's theorem is an immediate consequence of the definition

For a nodal fiber F of type I1, which is irreducible, we have the function field of rational functions on F, which may be identified with the field of rational functions on p1 The non-zero elements of this field form a multiplicative group, which has as a subgroup K; the group of functions which are regular and nonzero at the node This group K: comes equipped with a divisor function to the group of divisors Div(F 8m) supported on the smooth part F sm of F Since F 8m is a group, there is a natural map ~:Div(FSm)--*F sm which takes a formal sum of points of F sm to the actual sum in the group It is easy to check that Abel's theorem holds: a divisor DEDiv(F sm) is the divisor of a function f E E if and only if deg(D)=0 and ~(D)=0 in the group law of F srn

We will now extend this to fibers of type Im with m>2 With this assumption there is no field of functions to employ, since the fiber F is no longer irreducible Thus we must find a multiplicative group of appropriate functions without the aid of an associated field of rational functions

For this we rely on the existence of a set of coordinates on the components of F, which are adapted nicely to both the group law on F 8m and to the elliptic surface on which F lies

Definition 6 1 Suppose that the singular fiber F (which is assumed to be of type Irn) has components Co, C1, , Cm-1, with the zero section So meeting Co and Cj meeting only Cj• for each j Let uj be an affine coordinate on Cj for each j The set {uj} of coordinates will be called a standard set of a~ne coordinates on F if the following conditions hold:

(a) For each j , uj=O at CjNCj_I and uj=oo a#~-~Cjf~Cj+l (b) The map a: C*• Z/ rn -*F 8m sending a pair (t, j) to the point u j = t in

component Cj of F is an isomorphism of groups (c) For each j , if we set v i = l / u i to be the atone coordinate on Cj near uj=oo,

then vj and Uj+l extend to coordinate functions Vj and Uj+I on the elliptic surface near the point CjNCj+I, such that Vj=0 defines Cj+I and Uj+I =0 defines Cj near this point

Proposition 6 2 Let F be a fiber of type Im on a smooth elliptic surface X Then a standard set of a]fine coordinates exist on F Moreover, given the ordering of the components, there are exactly m such standard sets of a~ne coordinates on F

Proof The existence of a standard set of atone coordinates on F is an immedi-

Torsmn sections of elliptic surfaces 127

ate consequence of the local tonc description of a smooth sermstable elliptic surface near a singular fiber of type Im given m [AMRT, Chapter One, Section 4]. Indeed, the standard set of affine coordinates m exactly the set of coordinates on the tonc cover described there, wlnch descend mcely to X

Note that by C a) and (b), the coordinate u0 on Co is deternnned: it must be 0 at CoNCm-1, co at CoNCI, and 1 at SoNCo. Similarly, each coordinate u 3 is determined by the point uj = 1, which must be one of the points of C~ of approprmte order. Thus there are exactly m possibilities for Ul By (b), once Ul is determined, so m u 3. It is easy to check that these m different possibilitms all give standard sets (once it LS known that one of them does). []

Note that if (u0, ..., urn-l} IS a standard set of atone coordinates on F , then any other standard set is of the form (u~=r for some m th root of unity

We can now define the group of functmns ]C which plays the role of ratmnal flmctmns on F A bit of notation m useful. Suppose tha t g(u) is a nonzero ratmnal flmctmn of u. Firstly, define no(g) to be the order of g at u=0 , and noo(g) to be the order of g at u=oo (which LS the order of g(1/v) at v=O). These integers are of course independent of the chome of affine coordinate u. Moreover, we have that g(u)/u "~~ has a finite value at u=O, and g(u)u '~~176 has a finite value at u=c~. Secondly, define co(g) to be the value of g(u)/u n~ at u=O, and define coo(g) to be the value of g(u)u '*~176 at u--co. These constants do depend on the choice of coordinate u; they are simply the lowest coefficient of the Laurent series for g expanded about u=O and u=oo.

If we write

e I

(6.3) g(u) = I- [ / l-I ~=i k=l

with ~, Az and #k nonzero, a n d / E Z , then

(6.4) no (g) = l

(6.5) noo (g) =f-e-l

(6.6) co(g) = a ( - 1 ) ~ + l I I , \ ~ / I I ~k, and ~ k

(6.7) coo(g)

Fix a standard set {u: } of a ~ n e coordinates on F Define/C to be the set of ra-tuples of nonzero rational flmctlons (go(Uo), gl (Ul) , ---, gra--1 (Urn--l)) satmfymg the foUowmg conditions:

(a) For each 3, n~(g~)--Fno(g~+l)----O.


(b) For each j, Coo(gj)=Co(gj+l) (c) m-1 Z =0 0(gj)=0 Note that condition (a) says that the functions {gj) have opposite orders at the

nodes of F, and condition (b) says that with respect to the standard set of affine coordinates, they have the same leading coefficients in their Laurent series at these nodes These are local conditions about each of the nodes The final condition (c) is a global condition, saying that the orders sum to zero upon going around the cycle of components We note that K: is a multiplicative group (the operation being defined component-wise), and that it does not depend on the particular standard set of affine coordinates used

These conditions are motivated by a notion of restriction of functions from the surface to the fiber Suppose that G is a rational function on X We may uniquely write its divisor, near the fiber F, as div(G)=H+V, where V is the '~ertical" part of the divisor consisting of linear combinations of components of F, and H is the "horizontal" part of the divisor consisting of linear combinations of multi-sections for the fibration map If one restricts to the general fiber near F, one only sees the contributions from H We want to make the following regularity assumption for the function G:

{*) No curve appearing in the horizontal part H passes through any node of F

Under this assumption, we see that the zeroes and poles of G, as we approach the singular fiber F, survive in the smooth part F 8'~ of F Therefore we have a chance of obtaining a limiting version of Abel's theorem

Take then such a rational function G, and let us define a "restriction" to the fiber F, which will be an element of the group ]C Fix a standard set of affine coordinates {uj ) on F, and also assume that we have normalized the base curve so that ~r=0 along F Write V=)-~j rjCj as the vertical part of div(G) For each j , consider the ratio G/~ rj; this is a rational function on X which does not have a zero or pole identically along Cj (since 7r has a zero of order one along F) Restricting this function to Cj gives a nonzero rational function gj (uj)

L e m m a 6 8 If G satisfies the regularity condition (*), then the m tuple (go, ,gin-l) lies in ]C

Proof Near the point CjNCj+I, we have local coordinates Uj+I and Vj on X as in the definition of a standard set of affane coordinates The curve Cj is defined locally by Uj+I =0, and Cj+I by Vj =0 Moreover the fibration map r is locally of the form ~r=Uj+IVj Hence near CjACj+I, we may write G as

G(Uj+I, Vi) = U~.~_I VjrJ+~ L(Uj+I, Vj)


where the condition {.} implies that L(0, 0)#0 With this notation we have g j ( v J = v~) + '-r j L(0, vj) and gj+l(uj+l)=u~+S~+~n(uj+l,0)

Thus we see that n~(gj)=rj+l-r j and no(gj+l)=rj-rj+l, proving that con dition (a) of the definition of/C is satisfied

We also have that coo(gj=co(gj+l)=L(O, 0), giving us condition (b) Finally, the sum ~ j no (gj) telescopes to 0, showing that condition (c) holds []

From this point of view, the definition of K:, though at first glance rather ad- hoc, is actually quite natural Motivated by the above, we call/C the function group of the singular fiber F

We have a divisor map div: K:--,Div(FS'~), sending an m-tuple (go, , gin-l) to the formal sum of the zeroes and poles of each gj, throwing away any part of the divisor at the nodes This map is a group homomorphism Recall that we also have a natural summation map @ from Div(F sin) to F sm Abel's theorem for F can be stated as follows

T h e o r e m 6 9 A divisor DEDiv(F sin) is the divisor of an element of IC if and only if deg(D)=O and @(D)=O

Proof Let (g_)EK:, and let D=div(g_) For each j , write

i=1 k=l

From the definition of K:, we must have

lj-%+l

i k

J

and

Summiug the first set of equations over j , we see that

d e g ( D ) = E ( e j - f j ) = O J

Multiplying the second set of equations over j , and applying the above, we have that

~-'S~..~) = 1 i I-Ik #k

1 3 0 P i c k M i r a n d a a n d P e t e r S t i l l e r

which shows that the C* part of the group element @(D) is trivial Finally, to show that the Z/m part of @(D) is trivial, we must show that ~"~.jj(ej-fj)=O mod m Writing e j - f j as t j+l- l j using the first equation, we see that this sum telescopes to

J J

which is 0 mod m by the third equation This completes the proof of the necessity of the conditions on D

We leave the sufficiency, which is equally elementary, to the reader []

We note that the only elements (g_) of/C which have div(g_)--0 are the constant elements, where gj--c for every j with c being a fixed nonzero complex number We thus have an exact sequence

v- d i v ~,-.,. / T~sra'~ 0--* C* -* A. --* tnv0t~ )~'~F"~"*O

where Div0(F 8m) is the group of divisors of degree 0 on F 8m

Example 6 10 Suppose F is a fiber of type Ira}, with a standard set of affine coordinates {uj} Fix a primitive mth root of unity r and an integer c~E[0,m-1], and let p be the point of Ca} with coordinate u a k : r We note that p is a point of order m in the group law of F 8rn, so that D=mp-mO_ is a divisor on F s'~ with deg(D)=0 and @(D)=0 (where 0 is the origin of the group law, i e , the point in Co with coordinate uo=l )

By Abel's theorem, there is an element (g)EK: such that div(g)=D If c~=0, this element (g) is

(uo - r go = (uo_l)r~, gj----1 f o r j # 0

If l < a < m - 1 , we have

go = ( u o _ l ) m ,

g~ =u~ -'~

~ (~(--'I]rl. m ) ,

g j =

for j = 1, , a k - 1 ,

and

for j = a k + l , , m k - 1


7 The l imit Weil pair ing on a degenera te elliptic curve

The Well pairing on a smooth elliptic curve E is defined as follows (see [Sfl, Chapter III, Section 8]) Let S and T be two points of order m on E Choose a rational function g on E with div(g)=[m]*(T)-[m]*(O_O.), where [m] denotes multi- plication by m Then the Weil pairing em on the m-torsion points of E is defined by

e.~(S, T) = g(Xr

for any X E E where both g(X(gS) and g(X) are defined and nonzero (We use @ as the group law in E to avoid confusion ) The existence of the rational function g relies solely on Abel's theorem for E, as does the fact that em has values in the group of mth roots of unity

In the previous section, we developed an Abel theorem for a singular fiber F of type Is on a smooth elliptic surface, by replacing the notion of the field of rational functions with the limit function group ~ This allows us to define in the same way a limit Weil pairing on the m-torsion points of F, which we also denote by em In this section we compute it

Since the em pairing is isotropic and skew-symmetric, it suffices to compute era(S, T) for generators S and T of the group of m-torsion points on F

Fix an integer m; in order that F have m 2 m-torsion points, F must be of type Imk for some k Since the Weft pairing is formally invariant under base change, we may assume k= 1 and F is of type Im Also fix a standard set of affine coordinates {uj} on F Let r be a primitive mth root of unity Let T be the m-torsion point of F which is in component Co having coordinate u0 =r Let S be the m torsion point of F which is in component C1 having coordinate Ul--1 These points S and T generate the group of m-torsion points of F

Let u=exp(2ri/m 2) so that ym=r Consider the point T~EF in component Co having coordinate uo=u; note that mT~=T, and [m]*(T)=~,R(T'~R ) where the sum is taken over all m-torsion points of F Similarly, [m]*(O)=~'~R(R )

The element (_g)e/C such that div(g_)=[m]*(T)-[m]*(O) is defined by

u7-r gJ(uJ) = r 7-1

Now choose an XECo with coordinate uo=x The point X ~ S is then the point in C1 with coordinate x Thus

i x m - r / Xm--r ~---r g(X~S)/g(X) = gl (x)/go (x) = r ~ _ - ~

- / Xm--1

This shows that era(S, T) =r -1


Let M(t, s)=tT(gsS be the general point of order m in F; note that M(t, s) is in component C, and has coordinate us =r Extending the calculation above using the bilinearity and skew-symmetry we obtain the following

P r o p o s i t i o n 7 1 the form

With the above notation, the limit Well pairing on F takes

e,~(M(tl, sl), M(t2, sz)) = ~t182-t2sl

Note that this limit Weil pairing is computed using Abel's theorem for the limit function group/C; since/C is independent of the choice of the standard set of arlene coordinates used, as are deg(D) and ~(D), we see that Abel's Theorem and the limit Well pairing are both wen-defined, independent of the choices made (The formula of the above Proposition also depends on the choice of the sections S and T; if we choose different sections, the formula may change, but the limit Weil pairing does not )

Moreover the limit Well pairing is indeed the limit of the usual Well pairing on the nearby smooth fibers of the elliptic surface This follows from the nature of the element (_g)E/C used in the computation above: each gj has degree 0 on the component where it is defined, and hence is the usual restriction of a rational function G on the elliptic surface This function G can be chosen so that, when restricted to the nearby smooth fibers, it is the function used to define the Well pairing there Therefore the limit Weil pairing is the limit of the usual Well pairing

8 The roo t -o f -un i ty and c o m p o n e n t n n m b e r re la t ionship v ia the Weil pai r ing

Finally, we want to point out that the duality between the root-of-unity results and the component number results, which were mentioned in Section 4 as being related to the canonical involution, can be expressed also in terms of the Weft pairing We will compute this Weil pairing on the singular fibers using the limit Weft pairing developed in the previous section

First let W be a torsion section of order m of an elliptic surface passing through the zero component Co of an Im fiber Let (=exp(21ri/m) be a primitive mth root of unity Assume that the Im fiber is given a standard set of arlene coordinates, such that the point WNCo has coordinate uo=~ a, with ( a , m ) = l In the notation of the last section, we have W=aT Let W* be the set of (local) sections Z such that era(W, Z) - - ( By the computation given in Proposition 7 1, we see that such a local section Z is one which passes through component b, where ab-1 rood m


Specializing to the case where m is an odd prime p, and using the root-of-unity numbers l and the component numbers k, we see the following:

(8 i) If kj (W) = 0 and lj (W) = a, then W* is the set of local sections Z

with ks(Z ) =a -1

Finally let us return to the modular surface 7r: Yl(p)----~Xl(p) and the quotient ~d: Ylr(p)-*Xl(p), as described in Section 5 In the above, set W--T, the universal section, and fix a cusp x of type /1 on XI(p) Assume that lx(T)=a for this cusp, that is, TACo is the point with coordinate ~a The section T r on the quotient is induced by the multisection of r which, by (5 1), has kx=a -1 Since the Weil pairing is invariant under isogeny, we see by (8 1) that T t is exactly the image, locally, of the set W* of sections pairing with T to give value r Alternatively, we may write

T ' = image of e~(T, _)-1(~)

134 Rick Miranda and Peter Stiller: Torsion sections of elliptic surfaces

References

[AMRT] ASH, A , MUMFORD, D , RAPOPORT, M and TAI Y , Smooth Compactifications of Locally Symmetric Varieties, Math Sci Press, Brooldine, Mass, 1975

[CP] Cox, D A and PARRY, W R , Torsion in elliptic curves over k(t), Compasitio Math 41 (1980), 337-354

[KKMS] KEMPF, G KNUDSEN, F MUMFORD, D and SAINT DONAT B , Toroidal Embeddings I, Lecture Notes in Math 339, Springer Verlag, Berlin-Heidel berg 1973

[K] KODAIRA K , On compact analytic surfaces II, III Ann of Math 78 (1963), 1-40

[M] MIRANDA, R , Component numbers for torsion sections ofsemistable elliptic sur faces, in Proceedings of the Conference on Classification of Algebraic Varieties L 'Aquila, Italy, May 1992, Contemporary Math (to appear)

[MP] MIRANDA, R and PERSSON, U , Torsion groups of elliptic surfaces, Compositio Math 72 (1989), 249-267

[Shm] SHIMURA, G , Arithmetic Theory of Automorphic Functions, Princeton Univ Press, Princeton, N J , 1971

[Shd] SHIODA, T , On elliptic modular surfaces, J Math Soc Japan 24 (1972), 20-59 [Sill SILVERMAN, J H , The Arithmetic Theory of Elliptic Curves Graduate Texts in

Mathematics 106, Springer Verlag, New York-Berlin, 1986

Received October 4, 1993 Rick Miranda Department of Mathematics Colorado State University Ft Collins, CO 80523 U S A email: miranda~riemann math colostate edu

Peter Stiller Department of Mathematics Texas A&M University College Station, TX 77843 3368 U S A email: stiller~alggeo tamu edu

Torsion sections of elliptic surfaces - Tsinghua...

Documents

Transcript of Torsion sections of elliptic surfaces - Tsinghua...